Bad science: 3D-DXA study results explained

In previous blogs I have tried to explain how 3D-DXA can not measure the cortical bone parameters despite claims to the contrary by DMS-Apelem, their distributors and by Galgo Medical who are now selling this software by the name: “3D-SHAPER”. So why do we see several studies presented at ASBMR1,2 and even in journal papers3,4 with seemingly sensible and significant cortical bone changes (other than from a misuse use of statistics)? The answer is simple but I fear difficult to grasp for those that do not fully understand the technology. I hope that this blog post will explain it adequately, but do not hesitate to contact me if it is not quite clear yet.

3D-DXA uses a statistical model of the 3D shape and density distribution of the femur bone. This model is built from the CT scans of many femur bones. In the case of Galgo Medical and DMS-Apelem this set of CT scans is predominantly of postmenopausal Spanish woman, which in itself gives some issues as explained here. The model is defined by points on the surface and density values inside of the bone from all of these CT scans. A dimensionality reduction technique, such as Principal Component Analysis is used to converts all this data to only a few parameters. Now, here is comes the important part. Only the most important parameters are used to define a bone model in the 3D-DXA software. These parameters define the most common variations in the shape and density distribution. The main variation of the density distribution is the overall density, which is directly linked to the overall cortical thickness (Which I have already show in the original publication on the 3D-DXA technology5). What this means is that when you increase the value of this parameter, it will increase the overall density but at the same time also the cortical density and the cortical thickness. To find a 3D reconstruction of the bone with 3D-DXA this parameter is changed until the densities of the projection of the model matches the densities in the DXA image (with a large margin of error). So, if you have a dataset of DXA images with a low BMD at baseline and a high BMD at followup, thus an increase in BMD, your 3D-DXA analysis will also produce an increase in volumetric density, cortical density and cortical thickness. 3D-DXA does not measure any changes, it simply reflects the changes in BMD in the DXA image by the corresponding changes in the volumetric BMD as defined by the model. While the results of various studies using 3D-DXA might seem sensible, they do not reflect the actual changes in the population that was examined. They reflect the “normal” relationship between the cortical parameters and the overall density of the femur (in Spanish postmenopausal women). Let’s try to make this clear with our bicycle example again.


Say, we have data on the number of bicycles in Spain and France as well as the number of bicycle thefts. We can then built a model (red line) of this data, which can predict the number of thefts based on the number of bicycles a country has. The model makes perfect sense. More bicycles means more bicycle thefts. So far so good. Now, lets apply this model to a study where we want to compare the number of bicycles thefts between England and Holland. We collect data on the number of bicycles in each country and then apply our model to find the number of bicycle thefts (blue dots). When comparing the number of thefts we can see significant more thefts in Holland compared to England. It makes perfect sense since Holland has many more bicycles. But wait. When we look at the actual data on bicycle thefts we see only a small difference between England and Holland. The model was never measuring the number of thefts, it was simply estimating it based on the incidence in Spain and France. Perhaps France simply has a relatively larger percentage of thieves.

Galgo Medical consistently makes statements in the conclusion of their publications that the sensibility of the results indicate that 3D-DXA can be used for population studies, this of-course is entirely false. You will never see an increase in BMD with anything else than an increase in volumetric BMD, cortical BMD and cortical thickness. You will never see results from 3D-DXA with regions of significant increase in cortical thickness and a decrease in cortical density, which have been shown in studies from CT. You will never see a study using 3D-DXA showing significant cortical increases in one region and decreases in another, as has been seen in CT scans.

Many clinicians in the field of bone research are tempted with the prospects of playing with the big boys who have access to CT scans and trust these seemingly legitimate companies with the details of this technology. However, I believe that as a clinician you should fully understand the tools that you are using and the analysis that you are performing. If not, you will inevitably misinterpret the results and disseminate misleading studies, despite your best intentions.

  1. AM GALICH, L HUMBERT, R WINZENRIETH, L MAFFEI, V PREMROU, A FRIGERI, E VEGA, Analyzing the cortical and trabecular bone of the femur of patients with vertebral fractures by 3D-DXA. ASBMR 2017 Annual Meeting
  2. Mirella Lopez Picazo, Ludovic Humbert, Miguel Angel Gonzalez Ballester, Luis Del Rio, Renaud Winzenrieth, Silvana Di Gregorio, Changes in volumetric BMD and cortical thickness measured by 3D-DXA in the lumbar spine after 24 months of Denosumab treatment. ASBMR 2017 Annual Meeting
  3. Gifre L, Humbert L, Muxi A, Del Rio L, Vidal J, Portell E, Monegal A, Guañabens N, Peris P. Analysis of the evolution of cortical and trabecular bone compartments in the proximal femur after spinal cord injury by 3D-DXA. Osteoporos Int. 2017 Oct 17.
  4. Orduna G, Humbert L, Fonolla R, Romera J, Cos ML, Rial A, Nogués X, Diez-Perez A, Mellibovsky L. Cortical and Trabecular Bone Analysis of Patients With High Bone Mass From the Barcelona Osteoporosis Cohort Using 3-Dimensional Dual-Energy X-ray Absorptiometry: A Case-Control Study. J Clin Densitom. 2017 Jun 22.
  5. T. Whitmarsh, et al., Reconstructing the 3D shape and bone mineral density distribution of the proximal femur from dual-energy X-ray absorptiometry, IEEE Transactions on Medical Imaging, vol. 30(12), pp. 2101-14, 2011.